Web Search and Data Mining

Web Search and Data Mining

Lecture 4

Adapted from Manning, Raghavan and Schuetze

Recap of the last lecture

MapReduce and distributed indexing Scoring documents: linear comb/zone weighting tfidf term weighting and vector spaces Derivation of idf

This lecture

Vector space models Dimension reduction: random projection Review of linear algebra Latent semantic indexing (LSI)

Documents as vectors

At the end of Lecture 3 we said: Each doc d can now be viewed as a vector of

wfidf values, one component for each term So we have a vector space

terms are axes docs live in this space Dimension is usually very large

Why turn docs into vectors?

First application: Query-by-example Given a doc d, find others “like” it.

Now that d is a vector, find vectors (docs) “near” it.

Natural setting for bag of words model Dimension reduction

Intuition

Postulate: Documents that are “close together” in the vector space talk about the same things.

t1

d2

d1

d3

d4

d5

t3

t2

θ

φ

Measuring Document Similarity

Idea: Distance between d1 and d2 is the length of the vector |d1 – d2|. Euclidean distance

Why is this not a great idea? We still haven’t dealt with the issue of length

normalization Short documents would be more similar to each

other by virtue of length, not topic However, we can implicitly normalize by looking

at angles instead

Cosine similarity

Distance between vectors d1 and d2 captured by the cosine of the angle x between them.

Note – this is similarity, not distance No triangle inequality for similarity.

t 1

d 2

d 1

t 3

t 2

θ

Cosine similarity

A vector can be normalized (given a length of 1) by dividing each of its components by its length – here we use the L2 norm

This maps vectors onto the unit sphere:

Then,

Longer documents don’t get more weight

11 ,

n

i jij wd

i ix2

2x

Cosine similarity

Cosine of angle between two vectors The denominator involves the lengths of the

vectors.

n

i ki

n

i ji

n

i kiji

kj

kjkj

ww

ww

dd

ddddsim

1

2,1

2,

1 ,,),(

Normalization

Queries in the vector space model

Central idea: the query as a vector: We regard the query as short document We return the documents ranked by the

closeness of their vectors to the query, also represented as a vector.

Note that dq is very sparse!

n

i qi

n

i ji

n

i qiji

qj

qjqj

ww

ww

dd

ddddsim

1

2,1

2,

1 ,,),(

Dimensionality reduction

What if we could take our vectors and “pack” them into fewer dimensions (say 50,000100) while preserving distances?

(Well, almost.) Speeds up cosine computations.

Many possibilities including, Random projection. “Latent semantic indexing”.

Random projection onto k<<m axes

Choose a random direction x1 in the vector space.

For i = 2 to k, Choose a random direction xi that is orthogonal to

x1, x2, … xi–1.

Project each document vector into the subspace spanned by {x1, x2, …, xk}.

E.g., from 3 to 2 dimensions

d2

d1

x1

t 3

x2

t 2

t 1

x1

x2d2

d1

x1 is a random direction in (t1,t2,t3) space.x2 is chosen randomly but orthogonal to x1.

Dot product of x1 and x2 is zero.

Guarantee

With high probability, relative distances are (approximately) preserved by projection.

Computing the random projection

Projecting n vectors from m dimensions down to k dimensions: Start with m n matrix of terms docs, A. Find random k m orthogonal projection matrix

R. Compute matrix product W = R A.

jth column of W is the vector corresponding to doc j, but now in k << m dimensions.

Cost of computation

This takes a total of kmn multiplications. Expensive – see Resources for ways to do

essentially the same thing, quicker. Other variations, using sparse random matrix,

entries of R from {-1, 0, 1} with probabilities

{1/6, 2/3, 1/6}.

Why?

Latent semantic indexing (LSI)

Another technique for dimension reduction Random projection was data-independent LSI on the other hand is data-dependent

Eliminate redundant axes Pull together “related” axes – hopefully

car and automobile

Linear Algebra Background

Eigenvalues & Eigenvectors

Eigenvectors (for a square mm matrix S)

How many eigenvalues are there at most?

only has a non-zero solution if

this is a m-th order equation in λ which can have at most m distinct solutions (roots of the characteristic polynomial) – can be complex even though S is real.

eigenvalue(right) eigenvector

Example

Eigenvalues & Eigenvectors

0 and , 2121}2,1{}2,1{}2,1{ vvvSv

For symmetric matrices, eigenvectors for distincteigenvalues are orthogonal

TSS and 0 if ,complex for IS

All eigenvalues of a real symmetric matrix are real.

0vSv if then ,0, Swww Tn

All eigenvalues of a positive semidefinite matrix

are non-negative

Example

Let

Then

The eigenvalues are 1 and 3 (nonnegative, real). The eigenvectors are orthogonal (and real):

21

12S

.01)2(21

12 2

IS

1

1

1

1

Real, symmetric.

Plug in these values and solve for eigenvectors.

Let be a square matrix with m linearly independent eigenvectors (a “non-defective” matrix)

Theorem: Exists an eigen decomposition

(cf. matrix diagonalization theorem)

Columns of U are eigenvectors of S

Diagonal elements of are eigenvalues of

Eigen/diagonal Decomposition

diagonal

Unique for

distinct

eigen-values

Diagonal decomposition: why/how

nvvU ...1Let U have the eigenvectors as columns:

n

nnnn vvvvvvSSU

............

1

1111

Then, SU can be written

And S=UU–1.

Thus SU=U, or U–1SU=

Diagonal decomposition - example

Recall .3,1;21

1221

S

The eigenvectors and form

1

1

1

1

11

11U

Inverting, we have

2/12/1

2/12/11U

Then, S=UU–1 =

2/12/1

2/12/1

30

01

11

11

RecallUU–1 =1.

Example continued

Let’s divide U (and multiply U–1) by 2

2/12/1

2/12/1

30

01

2/12/1

2/12/1Then, S=

Q (Q-1= QT )

Why? Stay tuned …

If is a symmetric matrix:

Theorem: Exists a (unique) eigen

decomposition

where Q is orthogonal: Q-1= QT

Columns of Q are normalized eigenvectors

Columns are orthogonal.

(everything is real)

Symmetric Eigen Decomposition

TQQS

Time out!

I came to this class to learn about text retrieval and mining, not have my linear algebra past dredged up again … But if you want to dredge, Strang’s Applied

Mathematics is a good place to start. What do these matrices have to do with text? Recall m n term-document matrices … But everything so far needs square matrices – so

…

Singular Value Decomposition

TVUA

mm mn V is nn

For an m n matrix A of rank r there exists a factorization(Singular Value Decomposition = SVD) as follows:

The columns of U are orthogonal eigenvectors of AAT.

The columns of V are orthogonal eigenvectors of ATA.

ii

rdiag ...1 Singular values.

Eigenvalues 1 … r of AAT are the eigenvalues of ATA.

Singular Value Decomposition

Illustration of SVD dimensions and sparseness

SVD example

Let

01

10

11

A

Thus m=3, n=2. Its SVD is

2/12/1

2/12/1

00

30

01

3/16/12/1

3/16/12/1

3/16/20

Typically, the singular values arranged in decreasing order.

SVD can be used to compute optimal low-rank approximations.

Approximation problem: Find Ak of rank k such that

Ak and X are both mn matrices.

Typically, want k << r.

Low-rank Approximation

Frobenius normFkXrankX

k XAA

min)(:

Solution via SVD

Low-rank Approximation

set smallest r-ksingular values to zero

Tkk VUA )0,...,0,,...,(diag 1

column notation: sum of rank 1 matrices

Tii

k

i ik vuA

1

k

Approximation error

How good (bad) is this approximation? It’s the best possible, measured by the Frobenius

norm of the error:

where the i are ordered such that i i+1.

Suggests why Frobenius error drops as k increased.

1)(:

min

kFkFkXrankX

AAXA

SVD Low-rank approximation

Whereas the term-doc matrix A may have m=50000, n=10 million (and rank close to 50000)

We can construct an approximation A100 with rank 100. Of all rank 100 matrices, it would have the lowest

Frobenius error. Great … but why would we?? Answer: Latent Semantic Indexing

C. Eckart, G. Young, The approximation of a matrix by another of lower rank. Psychometrika, 1, 211-218, 1936.

Latent Semantic Analysis via SVD

What it is

From term-doc matrix A, we compute the approximation Ak.

There is a row for each term and a column for each doc in Ak

Thus docs live in a space of k<<r dimensions These dimensions are not the original

axes But why?

Vector Space Model: Pros

Automatic selection of index terms Partial matching of queries and documents

(dealing with the case where no document contains all search terms)

Ranking according to similarity score (dealing with large result sets)

Term weighting schemes (improves retrieval performance)

Various extensions Document clustering Relevance feedback (modifying query vector)

Geometric foundation

Problems with Lexical Semantics

Ambiguity and association in natural language Polysemy: Words often have a multitude

of meanings and different types of usage (more severe in very heterogeneous collections).

The vector space model is unable to discriminate between different meanings of the same word.

Problems with Lexical Semantics

Synonymy: Different terms may have identical or a similar meaning (weaker: words indicating the same topic).

No associations between words are made in the vector space representation.

Latent Semantic Indexing (LSI)

Perform a low-rank approximation of document-term matrix (typical rank 100-300)

General idea Map documents (and terms) to a low-

dimensional representation. Design a mapping such that the low-dimensional

space reflects semantic associations (latent semantic space).

Compute document similarity based on the inner product in this latent semantic space

Goals of LSI

Similar terms map to similar location in low dimensional space

Noise reduction by dimension reduction

Latent Semantic Analysis

Latent semantic space: illustrating example

courtesy of Susan Dumais

Performing the maps

Each row and column of A gets mapped into the k-dimensional LSI space, by the SVD.

Claim – this is not only the mapping with the best (Frobenius error) approximation to A, but in fact improves retrieval.

A query q is also mapped into this space, by

Query NOT a sparse vector.

1 kkT

k Uqq

Empirical evidence

Experiments on TREC 1/2/3 – Dumais Lanczos SVD code (available on netlib)

due to Berry used in these expts Running times of ~ one day on tens of

thousands of docs (old data) Dimensions – various values 250-350

reported (Under 200 reported unsatisfactory)

Generally expect recall to improve – what about precision?

Empirical evidence

Precision at or above median TREC precision Top scorer on almost 20% of TREC topics

Slightly better on average than straight vector spaces

Effect of dimensionality: Dimensions Precision

250 0.367

300 0.371

346 0.374

Some wild extrapolation

The “dimensionality” of a corpus is the number of distinct topics represented in it.

More mathematical wild extrapolation: if A has a rank k approximation of low

Frobenius error, then there are no more than k distinct topics in the corpus. (Latent semantic indexing: A probabilistic analysis,'' )

LSI has many other applications

In many settings in pattern recognition and retrieval, we have a feature-object matrix. For text, the terms are features and the docs are

objects. Could be opinions and users … This matrix may be redundant in dimensionality. Can work with low-rank approximation. If entries are missing (e.g., users’ opinions), can

recover if dimensionality is low. Powerful general analytical technique

Close, principled analog to clustering methods.